U.S. Treasury's machine learning AI prevents and recovers over $4 billion in fraudulent payments in fiscal year 2024

Background

The Treasury disburses trillions of dollars annually and faces an accelerating fraud environment: online payment fraud is projected to cumulatively surpass $362 billion by 2028 (Juniper Research). After pandemic-era surges in fraud, the OPI began deploying machine learning AI starting in late 2022 to move from reactive to proactive fraud detection.

What Was Implemented

• Expanded risk-based screening across the payments portfolio
• Machine learning AI for identifying and prioritizing high-risk transactions
• Dedicated ML tools for expediting detection of Treasury check fraud
• Data-sharing partnership with the Department of Labor (announced May 2024) to provide state unemployment agencies access to Do Not Pay Working System data

Results

In FY2024, the Treasury's ML-enhanced fraud detection contributed to preventing or recovering $4 billion+ total—a 6x increase over FY2023's $652.7 million. Machine learning specifically drove $1 billion in check fraud recovery . Risk-based screening alone prevented $500 million ; high-risk transaction prioritization prevented $2.5 billion ; payment schedule efficiencies prevented $180 million . Deputy Secretary Adeyemo attributed the gains to "dedicated efforts" by OPI to enhance fraud prevention capabilities.

Lessons

• Machine learning applied to high-volume government payment data demonstrates the scalability of AI-driven claim/fraud detection well beyond commercial settings
• Layered detection (risk-based screening + ML check fraud tools + payment schedule optimization) drives outsized results vs. any single method
• Data-sharing partnerships (Treasury/Labor example) amplify ML performance by enriching detection data
• A sixfold year-over-year improvement demonstrates rapid ROI expansion once ML infrastructure is in place